Re: [HACKERS] Runtime Partition Pruning

[Jesper Pedersen]

Hi David,

On 03/31/2018 09:52 AM, David Rowley wrote:

I've attached a new version of the patch.  I'm now at v18 after having
some versions of the patch that I didn't release which were based on
various versions of Amit's faster partition pruning patch.

Thank you for the updated patch set !

I have tested this together with Amit's v46 patch.



The standard table case (std.sql) gives:

 generic_cost = 8.4525
 avg_custom_cost = 13.4525
 total_custom_cost = 67.2625

whereas the 64 hash partition case (hash.sql) gives:

 generic_cost = 540.32
 avg_custom_cost = 175.9425
 total_custom_cost = 879.7125

I tested with pgbench -M prepared -f select.sql.



***************
*** 642,648 ****
  ---------------------+---+---+-----
   mlparted_tab_part1  | 1 | a |
   mlparted_tab_part2a | 2 | a |
!  mlparted_tab_part2b | 2 | b | xxx
   mlparted_tab_part3  | 3 | a | xxx
  (4 rows)

--- 642,648 ----
  ---------------------+---+---+-----
   mlparted_tab_part1  | 1 | a |
   mlparted_tab_part2a | 2 | a |
!  mlparted_tab_part2b | 2 | b |
   mlparted_tab_part3  | 3 | a | xxx
  (4 rows)

I'll spend some more time tomorrow.

Thanks for working on this !

Best regards,
 Jesper

The attached case doesn't trigger a generic plan, so basically all time is spent in GetCachedPlan. Also, I'm seeing a regression for check-world in src/test/regress/results/inherit.out

hash.sql
Description: application/sql

select.sql
Description: application/sql

std.sql
Description: application/sql